VGaussian: Dense Mapping of Indoor Drone Based on 3D Gaussian

Traditional SLAM(Simultaneous Localization and Mapping) mapping is difficult to preserve the color information of the scene and has holes, which poses a huge challenge to the authenticity of scene reconstruction. This article proposes a new framework that combines SLAM and radiation field for better mapping results. It combines a robust and multifunctional binocular vision inertial state estimator with 3D Gaussian to achieve robust localization and efficient mapping in moving real-world environments, effectively improving the problem of map holes and synthesizing color images with new perspectives. By deploying this method on quadcopter drones, the performance of our proposed method was validated in real-world experiments and compared with other advanced SLAM algorithms. The results showed that integrating binocular vision inertial state estimator with 3D Gaussian can achieve more realistic and realistic mapping effects.